DOCSLIB.ORG

Lecture 31: Introduction to Parallel I/O

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Lecture 31: Introduction to Parallel I/O Lecture 31: Introduction to Parallel I/O William Gropp www.cs.illinois.edu/~wgropp I/O and File Systems • Most applications need persistent storage ♦ Typical to store persistent data in files, accessed through input/output (I/O) features of programming language and runtime ♦ Dominant implementation for persistent storage is magnetic disks • Tape also used for higher capacity • Semiconductor and other technologies used for higher performance/lower power (e.g., FLASH) 2 Performance Parameters • Magnetic disks performance ♦ Latency 2-10 ms (time it take the disk to spin under the read/write head) • 1,000x slower than internode communication • 10,000,000x slower than processor core ♦ Bandwidth over 100MB/sec • But only approached for large transfers • Performance sensitive to exact usage pattern • Most I/O benchmarks are not representative of user I/O patterns ♦ At most, useful to set a performance goal • Most performance solutions aggregate operations in some way 3 Files and File Systems • A file is just an ordered collection of bytes • A file system manages collections of files and properties of the files ♦ Size ♦ Access restrictions ♦ Quotas ♦ Reading and writing data • File systems differ in the services they provide and the semantics of data access and update 4 Two Abstract I/O Models • Typical user view • A more accurate model Write a(1:10) Write a(1:10) Node Runtime 10 words Buffer 1Mblock 1 MB 5 Notes on the Abstract I/O Models • Leaves out caching ♦ Client (node) side can be hard (more later) • Leaves out metadata operations ♦ Does user still have access to the file? ♦ What is the date of the last access or change? ♦ How big is the file? ♦ Where is the file located? 6 Understanding Achieved File System Performance • Hardware operations don’t match programmer model ♦ Large block reads ♦ Writes are read-modify-write • Semantics don’t match most programmer’s model ♦ POSIX semantics for read/write to a file is similar to sequential consistency ♦ Makes it extremely hard to implement client- side caching, metadata updates • How much does this impact performance? 7 Few Applications Get Even 1% of I/O Performance Applications' Max Throughput platform Edison Intrepid Mira 1 TB/s 1 GB/s 1 MB/s I/O ThroughputI/O 1 KB/s 0 25% 50% 75% 100% Applications A Multiplatform Study of I/O Behavior on Petascale Supercomputers, Luu, Winslett, Gropp, Ross, Carns, Harms, Prabhat, Byna, Yao. HPDC’15, to appear 8 POSIX Read/Write Semantics • Once a write completes (at any process), any read, from any other process, must see that write • Basic operations have requirements that are often not understood and can impact performance ♦ In other words, few applications in scientific computing require this strong ordering of read and write operations between processes, but it impacts the performance and stability of the file system for everyone 9 POSIX Read and Write • Read and Write are atomic • No assumption on the number of processes (or their relationship to each other) that have a file open for reading and writing • Consider this case • Process 1 Process 2 read a … … write b read b 10 POSIX Read and Write • Reading a large block containing both a and b (Caching data) and using that data to perform the second read without going back to the original file is incorrect • This requirement of read/write results in the over specification of the interface in many applications codes ♦ Few applications require strong synchronization of read/write • There are similar mismatches with other operations… 11 Open • User’s model is that this gets a file descriptor and (perhaps) initializes local buffering • Should be fast and (more importantly) scalable operation • Problem: no Unix (or POSIX) interface for “exclusive access open” ♦ File system must assume that other processes will open the file ♦ Since the file system doesn’t know anything about a parallel program, there’s no case for when a unique job (rather than process) opens the file 12 Close • User’s model is that this flushes the last data written to disk (if they think about that) and relinquishes the file descriptor • When is data written out to disk? ♦ On close? ♦ Sometime later? ♦ Never? • Example: ♦ Unused physical memory pages used as disk cache (at the server) ♦ Combined with Uninterruptible Power Supply, data may never appear on disk ♦ Makes it extremely hard to write a robust I/O benchmark 13 Seek • User’s model is that this assigns the given location to a variable and takes about 1 nanosecond • Changes position in file for “next” read • May interact with implementation to cause data to flush data to disk (clear all caches) ♦ In that case very expensive, particularly when multiple processes are seeking into the same file 14 Read/Fread • Users expect read (unbuffered) to be faster than fread (buffered) (rule: buffering is bad, particularly when done by the user) ♦ Reverse true for short data (often by several orders of magnitude) ♦ User thinks reason is “System calls are expensive” ♦ Real culprit is atomic nature of read • Fortran since F77 requires unique open (Section 12.3.2, lines 44-45) ♦ Fortran spec designed for performance ♦ Still doesn’t help for parallel programs 15 Write • Users expect write to simply update that part of the file ♦ Often expect buffering (fwrite) even with write, or caching by the runtime • Users expect to get good performance with large writes ♦ E.g., write more than 1MB ♦ But efficient writes must be aligned with hardware/file system blocks • Unaligned writes can be much slower • May use general data path instead of special write-full-block path 16 What Does This Mean For I/O From Parallel Programs? • Many ways to organize I/O ♦ Should be considered in the context of the entire application workflow, not just one program • Several natural choices ♦ One file per program • May match workflow, other tools ♦ One file per process • Avoids performance and correctness bugs in the File system ♦ One file per node/row/rack/… • Workarounds for performance and correctness issues 17 Non-Parallel I/O • Non-parallel • Performance worse than sequential • Often legacy from before application was parallelized 18 Independent Parallel I/O • Each process writes to a separate file • Pro: parallelism • Con: lots of small files to manage • Either legacy from before MPI or done to address file system issues 19 Parallel I/O – Single File • Parallel • Performance can be great, good, bad, or terrible (even worse than sequential) • Depends on correct implementation of concurrent updates in file (all too rare) 20 Other Approaches • File systems with different consistency semantics ♦ NFS (esp version 3) – essentially no consistency; safe for serial access only ♦ PVFS – defines non-overlapping writes ♦ HDFS – Parallel access to immutable files • Other models of persistent data objects besides files ♦ Databases ♦ Object stores 21 Other Approaches: Hardware • The large performance gap between disks and memory makes performance difficult • New memory designs offering intermediate performance and cost ♦ Non-volatile RAM (NVRAM) ♦ Burst buffers (smooth out I/O performance demands with buffer to absorb writes) • Note useful only if POSIX semantics abandoned 22 Asking the Right Question • Do you want Unix or Fortran I/O? ♦ Even with a significant performance penalty? • Do you want to change your program? ♦ Even to another portable version with faster performance? ♦ Not even for a factor of 40??? • User “requirements” can be misleading 23 Readings • Darshan: P. Carns, K. Harms, W. Allcock, C. Bacon, S. Lang, R. Latham, R. Ross, Understanding and improving computational science storage access through continuous characterization, ACM Trans. on Storage, 7(3):8, 2011 ♦ Darshan is a tool to observe I/O usage by applications in HPC; current state-of-the-art in gathering data on HPC application I/O behavior 24 .
Load more

Recommended publications
  • Copy on Write Based File Systems Performance Analysis and Implementation
    Copy On Write Based File Systems Performance Analysis And Implementation Sakis Kasampalis Kongens Lyngby 2010 IMM-MSC-2010-63 Technical University of Denmark Department Of Informatics Building 321, DK-2800 Kongens Lyngby, Denmark Phone +45 45253351, Fax +45 45882673 [email protected] www.imm.dtu.dk Abstract In this work I am focusing on Copy On Write based file systems. Copy On Write is used on modern file systems for providing (1) metadata and data consistency using transactional semantics, (2) cheap and instant backups using snapshots and clones. This thesis is divided into two main parts. The first part focuses on the design and performance of Copy On Write based file systems. Recent efforts aiming at creating a Copy On Write based file system are ZFS, Btrfs, ext3cow, Hammer, and LLFS. My work focuses only on ZFS and Btrfs, since they support the most advanced features. The main goals of ZFS and Btrfs are to offer a scalable, fault tolerant, and easy to administrate file system. I evaluate the performance and scalability of ZFS and Btrfs. The evaluation includes studying their design and testing their performance and scalability against a set of recommended file system benchmarks. Most computers are already based on multi-core and multiple processor architec- tures. Because of that, the need for using concurrent programming models has increased. Transactions can be very helpful for supporting concurrent program- ming models, which ensure that system updates are consistent. Unfortunately, the majority of operating systems and file systems either do not support trans- actions at all, or they simply do not expose them to the users.
    [Show full text]
  • The Linux Kernel Module Programming Guide
    The Linux Kernel Module Programming Guide Peter Jay Salzman Michael Burian Ori Pomerantz Copyright © 2001 Peter Jay Salzman 2007−05−18 ver 2.6.4 The Linux Kernel Module Programming Guide is a free book; you may reproduce and/or modify it under the terms of the Open Software License, version 1.1. You can obtain a copy of this license at http://opensource.org/licenses/osl.php. This book is distributed in the hope it will be useful, but without any warranty, without even the implied warranty of merchantability or fitness for a particular purpose. The author encourages wide distribution of this book for personal or commercial use, provided the above copyright notice remains intact and the method adheres to the provisions of the Open Software License. In summary, you may copy and distribute this book free of charge or for a profit. No explicit permission is required from the author for reproduction of this book in any medium, physical or electronic. Derivative works and translations of this document must be placed under the Open Software License, and the original copyright notice must remain intact. If you have contributed new material to this book, you must make the material and source code available for your revisions. Please make revisions and updates available directly to the document maintainer, Peter Jay Salzman <[email protected]>. This will allow for the merging of updates and provide consistent revisions to the Linux community. If you publish or distribute this book commercially, donations, royalties, and/or printed copies are greatly appreciated by the author and the Linux Documentation Project (LDP).
    [Show full text]
  • Use of Seek When Writing Or Reading Binary Files
    Title stata.com file — Read and write text and binary files Description Syntax Options Remarks and examples Stored results Reference Also see Description file is a programmer’s command and should not be confused with import delimited (see [D] import delimited), infile (see[ D] infile (free format) or[ D] infile (fixed format)), and infix (see[ D] infix (fixed format)), which are the usual ways that data are brought into Stata. file allows programmers to read and write both text and binary files, so file could be used to write a program to input data in some complicated situation, but that would be an arduous undertaking. Files are referred to by a file handle. When you open a file, you specify the file handle that you want to use; for example, in . file open myfile using example.txt, write myfile is the file handle for the file named example.txt. From that point on, you refer to the file by its handle. Thus . file write myfile "this is a test" _n would write the line “this is a test” (without the quotes) followed by a new line into the file, and . file close myfile would then close the file. You may have multiple files open at the same time, and you may access them in any order. 1 2 file — Read and write text and binary files Syntax Open file file open handle using filename , read j write j read write text j binary replace j append all Read file file read handle specs Write to file file write handle specs Change current location in file file seek handle query j tof j eof j # Set byte order of binary file file set handle byteorder hilo j lohi j 1 j 2 Close
    [Show full text]
  • File Storage Techniques in Labview
    File Storage Techniques in LabVIEW Starting with a set of data as if it were generated by a daq card reading two channels and 10 samples per channel, we end up with the following array: Note that the first radix is the channel increment, and the second radix is the sample number. We will use this data set for all the following examples. The first option is to send it directly to a spreadsheet file. spreadsheet save of 2D data.png The data is wired to the 2D array input and all the defaults are taken. This will ask for a file name when the program block is run, and create a file with data values, separated by tab characters, as follows: 1.000 2.000 3.000 4.000 5.000 6.000 7.000 8.000 9.000 10.000 10.000 20.000 30.000 40.000 50.000 60.000 70.000 80.000 90.000 100.000 Note that each value is in the format x.yyy, with the y's being zeros. The default format for the write to spreadsheet VI is "%.3f" which will generate a floating point number with 3 decimal places. If a number with higher decimal places is entered in the array, it would be truncated to three. Since the data file is created in row format, and what you really need if you are going to import it into excel, is column format. There are two ways to resolve this, the first is to set the transpose bit on the write function, and the second, is to add an array transpose, located in the array pallet.
    [Show full text]
  • System Calls System Calls
    System calls We will investigate several issues related to system calls. Read chapter 12 of the book Linux system call categories file management process management error handling note that these categories are loosely defined and much is behind included, e.g. communication. Why? 1 System calls File management system call hierarchy you may not see some topics as part of “file management”, e.g., sockets 2 System calls Process management system call hierarchy 3 System calls Error handling hierarchy 4 Error Handling Anything can fail! System calls are no exception Try to read a file that does not exist! Error number: errno every process contains a global variable errno errno is set to 0 when process is created when error occurs errno is set to a specific code associated with the error cause trying to open file that does not exist sets errno to 2 5 Error Handling error constants are defined in errno.h here are the first few of errno.h on OS X 10.6.4 #define EPERM 1 /* Operation not permitted */ #define ENOENT 2 /* No such file or directory */ #define ESRCH 3 /* No such process */ #define EINTR 4 /* Interrupted system call */ #define EIO 5 /* Input/output error */ #define ENXIO 6 /* Device not configured */ #define E2BIG 7 /* Argument list too long */ #define ENOEXEC 8 /* Exec format error */ #define EBADF 9 /* Bad file descriptor */ #define ECHILD 10 /* No child processes */ #define EDEADLK 11 /* Resource deadlock avoided */ 6 Error Handling common mistake for displaying errno from Linux errno man page: 7 Error Handling Description of the perror () system call.
    [Show full text]
  • Ext4 File System and Crash Consistency
    1 Ext4 file system and crash consistency Changwoo Min 2 Summary of last lectures • Tools: building, exploring, and debugging Linux kernel • Core kernel infrastructure • Process management & scheduling • Interrupt & interrupt handler • Kernel synchronization • Memory management • Virtual file system • Page cache and page fault 3 Today: ext4 file system and crash consistency • File system in Linux kernel • Design considerations of a file system • History of file system • On-disk structure of Ext4 • File operations • Crash consistency 4 File system in Linux kernel User space application (ex: cp) User-space Syscalls: open, read, write, etc. Kernel-space VFS: Virtual File System Filesystems ext4 FAT32 JFFS2 Block layer Hardware Embedded Hard disk USB drive flash 5 What is a file system fundamentally? int main(int argc, char *argv[]) { int fd; char buffer[4096]; struct stat_buf; DIR *dir; struct dirent *entry; /* 1. Path name -> inode mapping */ fd = open("/home/lkp/hello.c" , O_RDONLY); /* 2. File offset -> disk block address mapping */ pread(fd, buffer, sizeof(buffer), 0); /* 3. File meta data operation */ fstat(fd, &stat_buf); printf("file size = %d\n", stat_buf.st_size); /* 4. Directory operation */ dir = opendir("/home"); entry = readdir(dir); printf("dir = %s\n", entry->d_name); return 0; } 6 Why do we care EXT4 file system? • Most widely-deployed file system • Default file system of major Linux distributions • File system used in Google data center • Default file system of Android kernel • Follows the traditional file system design 7 History of file system design 8 UFS (Unix File System) • The original UNIX file system • Design by Dennis Ritche and Ken Thompson (1974) • The first Linux file system (ext) and Minix FS has a similar layout 9 UFS (Unix File System) • Performance problem of UFS (and the first Linux file system) • Especially, long seek time between an inode and data block 10 FFS (Fast File System) • The file system of BSD UNIX • Designed by Marshall Kirk McKusick, et al.
    [Show full text]
  • File Handling in Python
    hapter C File Handling in 2 Python There are many ways of trying to understand programs. People often rely too much on one way, which is called "debugging" and consists of running a partly- understood program to see if it does what you expected. Another way, which ML advocates, is to install some means of understanding in the very programs themselves. — Robin Milner In this Chapter » Introduction to Files » Types of Files » Opening and Closing a 2.1 INTRODUCTION TO FILES Text File We have so far created programs in Python that » Writing to a Text File accept the input, manipulate it and display the » Reading from a Text File output. But that output is available only during » Setting Offsets in a File execution of the program and input is to be entered through the keyboard. This is because the » Creating and Traversing a variables used in a program have a lifetime that Text File lasts till the time the program is under execution. » The Pickle Module What if we want to store the data that were input as well as the generated output permanently so that we can reuse it later? Usually, organisations would want to permanently store information about employees, inventory, sales, etc. to avoid repetitive tasks of entering the same data. Hence, data are stored permanently on secondary storage devices for reusability. We store Python programs written in script mode with a .py extension. Each program is stored on the secondary device as a file. Likewise, the data entered, and the output can be stored permanently into a file.
    [Show full text]
  • System Calls and Standard I/O
    System Calls and Standard I/O Professor Jennifer Rexford http://www.cs.princeton.edu/~jrex 1 Goals of Today’s Class • System calls o How a user process contacts the Operating System o For advanced services that may require special privilege • Standard I/O library o Generic I/O support for C programs o A smart wrapper around I/O-related system calls o Stream concept, line-by-line input, formatted output, ... 2 1 System Calls 3 Communicating With the OS User Process signals systems calls Operating System • System call o Request to the operating system to perform a task o … that the process does not have permission to perform • Signal o Asynchronous notification sent to a process … to notify the process of an event that has occurred o 4 2 Processor Modes • The OS must restrict what a user process can do o What instructions can execute o What portions of the address space are accessible • Supervisor mode (or kernel mode) o Can execute any instructions in the instruction set – Including halting the processor, changing mode bit, initiating I/O o Can access any memory location in the system – Including code and data in the OS address space • User mode o Restricted capabilities – Cannot execute privileged instructions – Cannot directly reference code or data in OS address space o Any such attempt results in a fatal “protection fault” – Instead, access OS code and data indirectly via system calls 5 Main Categories of System Calls • File system o Low-level file I/O o E.g., creat, open, read, write, lseek, close • Multi-tasking mechanisms o Process
    [Show full text]
  • Filesystem Considerations for Embedded Devices ELC2015 03/25/15
    Filesystem considerations for embedded devices ELC2015 03/25/15 Tristan Lelong Senior embedded software engineer Filesystem considerations ABSTRACT The goal of this presentation is to answer a question asked by several customers: which filesystem should you use within your embedded design’s eMMC/SDCard? These storage devices use a standard block interface, compatible with traditional filesystems, but constraints are not those of desktop PC environments. EXT2/3/4, BTRFS, F2FS are the first of many solutions which come to mind, but how do they all compare? Typical queries include performance, longevity, tools availability, support, and power loss robustness. This presentation will not dive into implementation details but will instead summarize provided answers with the help of various figures and meaningful test results. 2 TABLE OF CONTENTS 1. Introduction 2. Block devices 3. Available filesystems 4. Performances 5. Tools 6. Reliability 7. Conclusion Filesystem considerations ABOUT THE AUTHOR • Tristan Lelong • Embedded software engineer @ Adeneo Embedded • French, living in the Pacific northwest • Embedded software, free software, and Linux kernel enthusiast. 4 Introduction Filesystem considerations Introduction INTRODUCTION More and more embedded designs rely on smart memory chips rather than bare NAND or NOR. This presentation will start by describing: • Some context to help understand the differences between NAND and MMC • Some typical requirements found in embedded devices designs • Potential filesystems to use on MMC devices 6 Filesystem considerations Introduction INTRODUCTION Focus will then move to block filesystems. How they are supported, what feature do they advertise. To help understand how they compare, we will present some benchmarks and comparisons regarding: • Tools • Reliability • Performances 7 Block devices Filesystem considerations Block devices MMC, EMMC, SD CARD Vocabulary: • MMC: MultiMediaCard is a memory card unveiled in 1997 by SanDisk and Siemens based on NAND flash memory.
    [Show full text]
  • X86-64 Machine-Level Programming∗
    x86-64 Machine-Level Programming∗ Randal E. Bryant David R. O'Hallaron September 9, 2005 Intel’s IA32 instruction set architecture (ISA), colloquially known as “x86”, is the dominant instruction format for the world’s computers. IA32 is the platform of choice for most Windows and Linux machines. The ISA we use today was defined in 1985 with the introduction of the i386 microprocessor, extending the 16-bit instruction set defined by the original 8086 to 32 bits. Even though subsequent processor generations have introduced new instruction types and formats, many compilers, including GCC, have avoided using these features in the interest of maintaining backward compatibility. A shift is underway to a 64-bit version of the Intel instruction set. Originally developed by Advanced Micro Devices (AMD) and named x86-64, it is now supported by high end processors from AMD (who now call it AMD64) and by Intel, who refer to it as EM64T. Most people still refer to it as “x86-64,” and we follow this convention. Newer versions of Linux and GCC support this extension. In making this switch, the developers of GCC saw an opportunity to also make use of some of the instruction-set features that had been added in more recent generations of IA32 processors. This combination of new hardware and revised compiler makes x86-64 code substantially different in form and in performance than IA32 code. In creating the 64-bit extension, the AMD engineers also adopted some of the features found in reduced-instruction set computers (RISC) [7] that made them the favored targets for optimizing compilers.
    [Show full text]
  • The Third Extended File System with Copy-On-Write
    Limiting Liability in a Federally Compliant File System Zachary N. J. Peterson The Johns Hopkins University Hopkins Storage Systems Lab, Department of Computer Science Regulatory Requirements z Data Maintenance Acts & Regulations – HIPAA, GISRA, SOX, GLB – 4,000+ State and Federal Laws and Regulations with regards to storage z Audit Trail – creating a “chain of trust” – Files are versioned over time – Authenticated block sharing (copy-on-write) between versions. z Disk Encryption – Privacy and Confidentiality – Non-repudiation Hopkins Storage Systems Lab, Department of Computer Science Secure Deletion in a Regulatory Environment z Desire to limit liability when audited – Records that go out of audit scope do so forever – When a disk is subpoenaed old or irrelevant data are inaccessible z Existing Techniques – Secure overwrite [Gutmann] – File key disposal in disk encrypted systems [Boneh & Lipton] z Existing solutions don’t work well in block- versioning file systems Hopkins Storage Systems Lab, Department of Computer Science Technical Problems z Secure overwriting of noncontiguous data blocks is slow and inefficient – When versions share blocks, data to be overwritten may be noncontiguous z Cannot dispose file keys in a versioning file system – Blocks encrypted with a particular key need to be available in future versions z User space tools are inadequate – Can’t delete metadata – Can’t be interposed between file operations – Truncate may leak data – Difficult to be synchronous Hopkins Storage Systems Lab, Department of Computer Science
    [Show full text]
  • X86 Assembly Language: AT&T and Intel
    x86_16 real mode (or at least enough for cos318 project 1) Overview Preliminary information - How to find help The toolchain The machine If you only remember one thing: gcc -S the -S (capital S) flag causes gcc to ouput assembly. Preliminary Information Assembly can be hard Development strategies conquer risk: Write small test cases. Write functions, test each separately. Print diagnostics frequently. Think defensively! and the interweb is helpful too. The Interwebs as a resource. The internet offers much information that seems confusing or contradictory. How do you sort out information "in the wild?" Syntax There are (at least) two different syntaxes for x86 assembly language: AT&T and Intel. AT&T: opcodes have a suffix to denote data type, use sigils, and place the destination operand on the right. Intel: operands use a keyword to denote data type, no sigils, destination operand is leftmost. Example: AT&T vs Intel push %bp push bp mov %sp,%bp mov bp,sp sub $0x10,%sp sub sp,0x10 movw mov si,WORD 0x200b(%bx),%si PTR [bx+0x200b] mov $0x4006,%di mov di,0x4006 mov $0x0,%ax mov ax,0x0 call printf call printf leaveq leave retq ret In this class, use AT&T! Versions of the architecture x86 won't die. All backwards compatible. 8086 -> 16bit, Real 80386 / ia32 -> 32bit, Protected x86_64 -> 64bit, Protected If you find an example: For which architecture was it written? The Register Test If you see "%rax", then 64-bit code; else If you see "%eax", then 32-bit code; else You are looking at 16-bit code.
    [Show full text]